Beyond Static Parameters: Adaptive Simulated Annealing in Practice

Abstract

The North Sea Locks connect the saline Western Scheldt to the Terneuzen-Gent canal, requiring efficient scheduling of vessels. While current scheduling models minimise vessel delays, lockages also cause freshwater loss and salt intrusion into the canal, negatively affecting drinking water, agriculture, and the surrounding ecosystem. This thesis extends an existing simulated annealing scheduling model with water loss and salt intrusion objectives, and evaluates adaptive hyperparameter mechanisms to reduce the burden of empirical tuning. The results confirm a direct tradeoff between the environmental impact and vessel delays; incorporating environmental objectives yields significant reductions in freshwater loss and salt intrusion at the cost of increased vessel delays. Among the adaptive approaches, those guided by domain-specific information, such as tidal conditions and salinity levels, show performance improvements. General approaches from the literature, specifically the Modified Lam Annealing schedule and memory-based neighbourhood selection, do not transfer effectively to the North Sea Lock scheduling problem. These findings suggest that for domain-specific optimisation problems, problem-specific adaptive mechanisms outperform general-purpose ones, and that meaningful environmental gains in lock scheduling are achievable through optimisation alone.